All tracks

AI Foundations · Free lesson · 2 min read

How Models Think

A large language model does one thing astonishingly well: it predicts the next word. Trained on a huge amount of text, it learned the patterns of how language fits together. There is no database it looks things up in and no little reasoner inside — just a very good guess at what comes next, made one word at a time.

At its core, what is a language model actually doing when it answers you?

  1. Searching a database of stored answers
  2. Predicting likely next words from patterns it learned
  3. Following a fixed set of hand-written rules
  4. Asking a human to write the reply
Show the answer

Predicting likely next words from patterns it learned

It generates text one token at a time by predicting what is most likely to come next. That is why it can be fluent and wrong at the same time.

In one sentence, explain to a friend why an AI can give a confident answer that turns out to be false.

Write your own answer in the mobile app to get this evaluated. The web reader is read-only for now.

What a strong answer covers

Strong answers connect it to prediction: the model produces fluent, likely-sounding text rather than verified facts, so confidence reflects pattern-fit, not correctness.

Save your streak and keep going.

Create a free account to unlock the full AI Foundations track, get AI-evaluated feedback on your written answers, and build a streak that compounds.

An unhandled error has occurred. Reload 🗙

Rejoining the server...

Rejoin failed... trying again in seconds.

Failed to rejoin.
Please retry or reload the page.

The session has been paused by the server.

Failed to resume the session.
Please retry or reload the page.